论文部分内容阅读
大规模的红外光谱数据集中存在大量无关冗余的特征。针对这一问题,提出了一种动态赋权红外光谱特征选择算法(Dynamic Weight Infrared Spectrum Feature Selection Algorithm,MBDWFS)。该算法把对称不确定性度量标准与近似Markov Blanket相结合,以删除原始光谱数据集中无关冗余的特征,从而获取数据规模较小且最优的特征子集。通过与FCBF、ID_3和ReliefF三种经典特征选择算法的性能仿真对比试验,证明所提出的MBDWFS算法在整体分类性能上优于其他三种算法,用于红外光谱的物质分析领域时效果更好。
There is a large amount of irrelevant features in large-scale infrared spectral data sets. To solve this problem, a Dynamic Weight Infrared Spectrum Feature Selection Algorithm (MBDWFS) is proposed. The algorithm combines the symmetry uncertainty metric with the approximate Markov Blanket to remove the irrelevant features of the original spectral dataset and obtain the smaller and optimal subset of the features. By comparing with the performance simulation of three classic feature selection algorithms FCBF, ID_3 and ReliefF, it is proved that the proposed MBDWFS algorithm outperforms the other three algorithms in the overall classification performance and is better in the field of material analysis for infrared spectroscopy.